taskforce-loop-engineering 0.13.0 → 0.15.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
@@ -0,0 +1,41 @@
1
+ import { createHash, randomUUID } from 'node:crypto';
2
+ import { mkdir, readFile, readdir, rename, rm, stat, writeFile } from 'node:fs/promises';
3
+ import path from 'node:path';
4
+
5
+ export const EXECUTION_LEDGER_VERSION = 2;
6
+ const KINDS = new Set(['llm', 'tool', 'effect']);
7
+ const TERMINAL = new Set(['succeeded', 'failed', 'cancelled']);
8
+ const SENSITIVE = /(^|_)(secret|token|password|credential|api[_-]?key|private[_-]?key)(_|$)/i;
9
+ const canonicalValue = (value) => Array.isArray(value) ? value.map(canonicalValue) : value && typeof value === 'object' ? Object.fromEntries(Object.keys(value).sort().map((key) => [key, canonicalValue(value[key])])) : value;
10
+ const hash = (value) => createHash('sha256').update(typeof value === 'string' ? value : JSON.stringify(canonicalValue(value))).digest('hex');
11
+ export const stepInputFingerprint = (input) => hash(input);
12
+ function text(value, name) { if (typeof value !== 'string' || !value.trim()) throw new Error(`${name} must be a non-empty string.`); return value; }
13
+ function safeId(value, name = 'stepId') { text(value, name); if (!/^[a-zA-Z0-9][a-zA-Z0-9._:-]{0,299}$/.test(value)) throw new Error(`${name} contains unsafe characters or is too long.`); return value; }
14
+ function redact(value, key = '') { if (SENSITIVE.test(key)) return '[REDACTED]'; if (Array.isArray(value)) return value.map((item) => redact(item)); if (value && typeof value === 'object') return Object.fromEntries(Object.keys(value).sort().map((name) => [name, redact(value[name], name)])); return value; }
15
+ function locations(root, stepId) { const directory = path.join(path.resolve(root), 'runtime', 'loops', 'execution-ledger'); const key = hash(safeId(stepId)); return { file: path.join(directory, 'steps', `${key}.json`), lock: path.join(directory, 'locks', key) }; }
16
+ async function atomicWrite(file, value) { await mkdir(path.dirname(file), { recursive: true }); const temporary = `${file}.${process.pid}.${randomUUID()}.tmp`; await writeFile(temporary, `${JSON.stringify(value, null, 2)}\n`, { flag: 'wx', mode: 0o600 }); await rename(temporary, file); }
17
+ async function mutex(lock) { await mkdir(path.dirname(lock), { recursive: true }); const deadline = Date.now() + 5000; while (true) { try { await mkdir(lock); return; } catch (error) { if (error.code !== 'EEXIST') throw error; const info = await stat(lock).catch(() => null); if (info && Date.now() - info.mtimeMs > 30_000) await rm(lock, { recursive: true, force: true }); else if (Date.now() >= deadline) throw new Error('Timed out acquiring execution ledger mutex.'); else await new Promise((resolve) => setTimeout(resolve, 10)); } } }
18
+ async function mutate(root, stepId, operation) { const loc = locations(root, stepId); await mkdir(path.dirname(loc.file), { recursive: true }); await mutex(loc.lock); try { const record = await readFile(loc.file, 'utf8').then(JSON.parse).catch((error) => error.code === 'ENOENT' ? null : Promise.reject(error)); const result = await operation(record); if (result.write) await atomicWrite(loc.file, result.record); return result.output ?? result.record; } finally { await rm(loc.lock, { recursive: true, force: true }); } }
19
+ function event(record, type, data = {}) { const entry = { sequence: record.events.length + 1, type, at: new Date().toISOString(), ...redact(data), previous: record.events.at(-1)?.checksum ?? null }; entry.checksum = hash(entry); return entry; }
20
+ function assertFence(record, token) { if (record.status !== 'running' || record.lease?.fencing_token !== Number(token)) throw new Error('Stale or invalid fencing token.'); }
21
+
22
+ export async function registerStep(root, input) {
23
+ const stepId = safeId(input.stepId); const kind = text(input.kind, 'kind'); if (!KINDS.has(kind)) throw new Error(`Unsupported step kind: ${kind}`); const fingerprint = stepInputFingerprint(input.input);
24
+ return mutate(root, stepId, async (record) => {
25
+ if (record) { if (record.kind !== kind || record.input_fingerprint !== fingerprint) throw new Error('Replay divergence: step identity is bound to a different kind or input fingerprint.'); return { write: false, output: { created: false, reusable: record.status === 'succeeded' && record.policy.reusable, record } }; }
26
+ const now = new Date().toISOString(); const created = { schema: 'loop.execution_step', version: 2, step_id: stepId, kind, status: 'pending', input_fingerprint: fingerprint, input: redact(input.input), lineage: { run_id: input.runId ?? null, task_id: input.taskId ?? null, parent_step_id: input.parentStepId ?? null, replaces_step_id: input.replacesStepId ?? null }, policy: { reusable: input.reusable !== false, effect_key: kind === 'effect' ? safeId(input.effectKey ?? stepId, 'effectKey') : null, authorization_scope: input.authorizationScope ?? null }, attempt: 0, fencing_counter: 0, lease: null, checkpoints: [], outcome: null, evidence: [], reconciliation: null, created_at: now, updated_at: now, events: [] }; created.events.push(event(created, 'registered', { input_fingerprint: fingerprint })); return { write: true, record: created, output: { created: true, reusable: false, record: created } };
27
+ });
28
+ }
29
+ export async function inspectStep(root, stepId) { const loc = locations(root, stepId); return readFile(loc.file, 'utf8').then(JSON.parse).catch((error) => error.code === 'ENOENT' ? null : Promise.reject(error)); }
30
+ export async function claimStep(root, input) {
31
+ const leaseMs = Number(input.leaseMs ?? 60_000); if (!Number.isInteger(leaseMs) || leaseMs <= 0) throw new Error('leaseMs must be a positive integer.');
32
+ return mutate(root, input.stepId, async (record) => { if (!record) throw new Error('Step must be registered before claim.'); if (record.status === 'succeeded') return { write: false, output: { claimed: false, reason: record.policy.reusable ? 'reusable_outcome' : 'terminal', record } }; if (TERMINAL.has(record.status)) return { write: false, output: { claimed: false, reason: 'terminal', record } }; if (record.status === 'unknown') return { write: false, output: { claimed: false, reason: 'reconcile_required', record } }; const nowMs = Date.now(); if (record.status === 'running' && Date.parse(record.lease.expires_at) > nowMs) return { write: false, output: { claimed: false, reason: 'lease_active', record } }; if (record.status === 'running') { const at = new Date().toISOString(); const next = { ...record, status: 'unknown', lease: null, reconciliation: { required: true, reason: 'lease_expired_after_claim', at }, updated_at: at }; next.events = [...record.events, event(record, 'outcome_unknown', { reason: 'lease_expired_after_claim' })]; return { write: true, record: next, output: { claimed: false, reason: 'reconcile_required', record: next } }; } const token = record.fencing_counter + 1; const at = new Date().toISOString(); const next = { ...record, status: 'running', attempt: record.attempt + 1, fencing_counter: token, lease: { owner: text(input.owner, 'owner'), fencing_token: token, claimed_at: at, expires_at: new Date(nowMs + leaseMs).toISOString() }, updated_at: at }; next.events = [...record.events, event(record, 'claimed', { attempt: next.attempt, fencing_token: token, owner: input.owner })]; return { write: true, record: next, output: { claimed: true, fencingToken: token, attempt: next.attempt, record: next } }; });
33
+ }
34
+ export async function checkpointStep(root, input) { return mutate(root, input.stepId, async (record) => { assertFence(record, input.fencingToken); const checkpoint = { sequence: record.checkpoints.length + 1, at: new Date().toISOString(), fingerprint: stepInputFingerprint(input.state), state: redact(input.state), evidence: redact(input.evidence ?? null) }; const next = { ...record, checkpoints: [...record.checkpoints, checkpoint], updated_at: checkpoint.at }; next.events = [...record.events, event(record, 'checkpointed', { sequence: checkpoint.sequence, fingerprint: checkpoint.fingerprint })]; return { write: true, record: next, output: checkpoint }; }); }
35
+ export async function completeStep(root, input) { return mutate(root, input.stepId, async (record) => { if (record?.status === 'succeeded') return { write: false, output: { completed: false, duplicate: true, record } }; assertFence(record, input.fencingToken); const at = new Date().toISOString(); const outcome = { status: 'succeeded', output_fingerprint: stepInputFingerprint(input.output), output: redact(input.output), at }; const next = { ...record, status: 'succeeded', lease: null, outcome, evidence: [...record.evidence, redact(input.evidence ?? {})], reconciliation: null, updated_at: at }; next.events = [...record.events, event(record, 'succeeded', { output_fingerprint: outcome.output_fingerprint })]; return { write: true, record: next, output: { completed: true, duplicate: false, record: next } }; }); }
36
+ export async function markStepUnknown(root, input) { return mutate(root, input.stepId, async (record) => { assertFence(record, input.fencingToken); const at = new Date().toISOString(); const next = { ...record, status: 'unknown', lease: null, reconciliation: { required: true, reason: input.reason ?? 'submit_outcome_unknown', at }, evidence: [...record.evidence, redact(input.evidence ?? {})], updated_at: at }; next.events = [...record.events, event(record, 'outcome_unknown', { reason: next.reconciliation.reason })]; return { write: true, record: next }; }); }
37
+ export async function reconcileStep(root, input) { return mutate(root, input.stepId, async (record) => { if (!record || record.status !== 'unknown') throw new Error('Only an unknown step can be reconciled.'); if (!['accepted', 'not_accepted'].includes(input.outcome)) throw new Error('Reconciliation outcome must be accepted or not_accepted.'); if (!input.evidence) throw new Error('Reconciliation requires evidence.'); const at = new Date().toISOString(); const evidence = redact(input.evidence); if (input.outcome === 'accepted') { const outcome = { status: 'succeeded', output_fingerprint: stepInputFingerprint(input.output), output: redact(input.output), reconciled: true, at }; const next = { ...record, status: 'succeeded', outcome, evidence: [...record.evidence, evidence], reconciliation: { required: false, outcome: 'accepted', at }, updated_at: at }; next.events = [...record.events, event(record, 'reconciled_accepted', { output_fingerprint: outcome.output_fingerprint })]; return { write: true, record: next }; } const next = { ...record, status: 'pending', lease: null, evidence: [...record.evidence, evidence], reconciliation: { required: false, outcome: 'not_accepted', at }, updated_at: at }; next.events = [...record.events, event(record, 'reconciled_not_accepted')]; return { write: true, record: next }; }); }
38
+ export async function replayDecision(root, input) { const record = await inspectStep(root, input.stepId); if (!record) return { action: 'register_and_execute', record: null }; if (record.kind !== input.kind || record.input_fingerprint !== stepInputFingerprint(input.input)) return { action: 'divergence', reason: 'input_fingerprint_mismatch', record }; if (record.status === 'succeeded' && record.policy.reusable) return { action: 'reuse', output: record.outcome.output, evidence: record.evidence, record }; if (record.status === 'unknown' || record.status === 'running') return { action: 'reconcile', record }; if (record.status === 'pending') return { action: 'execute', checkpoint: record.checkpoints.at(-1) ?? null, record }; return { action: 'terminal', record }; }
39
+ export async function listSteps(root) { const dir = path.join(path.resolve(root), 'runtime', 'loops', 'execution-ledger', 'steps'); const result = []; for (const name of (await readdir(dir).catch(() => [])).filter((name) => name.endsWith('.json')).sort()) result.push(JSON.parse(await readFile(path.join(dir, name), 'utf8'))); return result.sort((a, b) => a.step_id.localeCompare(b.step_id)); }
40
+ export async function migrateLegacyJournalStep(root, legacy) { return registerStep(root, { stepId: legacy.step_id ?? legacy.stepId ?? `legacy:${legacy.transactionId ?? hash(legacy).slice(0, 24)}`, kind: KINDS.has(legacy.kind) ? legacy.kind : 'tool', input: legacy.input ?? legacy.payload ?? legacy, taskId: legacy.task_id, runId: legacy.run_id, reusable: legacy.reusable !== false }); }
41
+ export async function migrateActionReservation(root, reservation) { const registered = await registerStep(root, { stepId: `effect:${reservation.idempotency_key}`, kind: 'effect', effectKey: reservation.idempotency_key, input: reservation.request, authorizationScope: reservation.authorization?.scope, reusable: true }); return { ...registered, legacy_state: reservation.state, migration: 'action_reservation_to_effect_protocol_v2' }; }
@@ -102,9 +102,23 @@ async function actions(root, warnings, nowMs) {
102
102
  return result.sort((a, b) => String(a.idempotency_key).localeCompare(String(b.idempotency_key)));
103
103
  }
104
104
 
105
+ async function executionSteps(root, warnings, nowMs) {
106
+ const result = [];
107
+ for (const file of await files(path.join(root, 'runtime', 'loops', 'execution-ledger', 'steps'))) {
108
+ const item = await json(file, warnings, root); if (!item) continue;
109
+ result.push(clean({ step_id: item.step_id, kind: item.kind, state: normalizedState(item, null, nowMs), status: item.status, input_fingerprint: item.input_fingerprint, attempt: item.attempt, lineage: item.lineage, lease: item.lease, checkpoint_count: item.checkpoints?.length ?? 0, outcome: item.outcome, evidence: item.evidence, reconciliation: item.reconciliation, updated_at: item.updated_at }));
110
+ }
111
+ return result.sort((a, b) => a.step_id.localeCompare(b.step_id));
112
+ }
113
+
114
+ async function productionEvidence(root, warnings) {
115
+ const file = path.join(root, '.production-evidence', 'public-summary.json');
116
+ try { return clean(await json(file, warnings, root)); } catch { return null; }
117
+ }
118
+
105
119
  async function legacyQueues(root, warnings, nowMs) {
106
120
  const loops = path.join(root, 'runtime', 'loops');
107
- const excluded = new Set(['control-plane', 'action-reservations', 'projects']);
121
+ const excluded = new Set(['control-plane', 'action-reservations', 'execution-ledger', 'projects']);
108
122
  const queues = [];
109
123
  for (const name of (await dirs(loops)).filter((item) => !excluded.has(item))) {
110
124
  const base = path.join(loops, name); const counts = Object.fromEntries([...STATES].map((state) => [state, 0])); const tasks = [];
@@ -138,13 +152,13 @@ export async function buildOperatorProjection(root, options = {}) {
138
152
  const resolved = path.resolve(root); const now = options.now ? new Date(options.now) : new Date();
139
153
  if (Number.isNaN(now.getTime())) throw new Error('Invalid projection time.');
140
154
  const warnings = []; const before = await stat(path.join(resolved, 'runtime', 'loops')).catch(() => null);
141
- const [control, reservations, queueList, projectList] = await Promise.all([controlPlane(resolved, warnings, now.getTime()), actions(resolved, warnings, now.getTime()), legacyQueues(resolved, warnings, now.getTime()), projects(resolved, warnings)]);
142
- const newest = [control.updated_at, ...control.todos.map((item) => item.updated_at), ...reservations.map((item) => item.updated_at), ...queueList.flatMap((queue) => queue.tasks.map((item) => item.updated_at))].filter(Boolean).sort().at(-1) ?? null;
155
+ const [control, reservations, steps, queueList, projectList, trustEvidence] = await Promise.all([controlPlane(resolved, warnings, now.getTime()), actions(resolved, warnings, now.getTime()), executionSteps(resolved, warnings, now.getTime()), legacyQueues(resolved, warnings, now.getTime()), projects(resolved, warnings), productionEvidence(resolved, warnings)]);
156
+ const newest = [control.updated_at, ...control.todos.map((item) => item.updated_at), ...reservations.map((item) => item.updated_at), ...steps.map((item) => item.updated_at), ...queueList.flatMap((queue) => queue.tasks.map((item) => item.updated_at))].filter(Boolean).sort().at(-1) ?? null;
143
157
  const counts = Object.fromEntries([...STATES].map((state) => [state, 0]));
144
158
  for (const item of [...control.todos, ...queueList.flatMap((queue) => queue.tasks)]) counts[item.state] += 1;
145
159
  const after = await stat(path.join(resolved, 'runtime', 'loops')).catch(() => null);
146
160
  if (before && after && before.mtimeMs !== after.mtimeMs) warnings.push({ code: 'concurrent_update', artifact: 'runtime/loops', message: 'Artifacts changed while the projection was read; refresh recommended.' });
147
- return clean({ schema_version: DASHBOARD_SCHEMA_VERSION, generated_at: now.toISOString(), source: { root: resolved, read_only: true, newest_artifact_at: newest, freshness_seconds: newest ? Math.max(0, Math.floor((now.getTime() - Date.parse(newest)) / 1000)) : null }, health: { status: warnings.length ? 'degraded' : 'ok', warnings }, overview: { counts, queue_count: queueList.length, project_count: projectList.length, todo_count: control.todos.length, action_count: reservations.length }, projects: projectList, queues: queueList, todos: control.todos, agents: control.agents, handoffs: control.handoffs, gates: control.todos.filter((item) => ['parked', 'waiting_for_human', 'waiting_for_external_condition', 'timed_out_or_escalated'].includes(item.state)).map((item) => ({ todo_id: item.id, state: item.state, gate: item.gate, next_action: item.next_action })), actions: reservations, cost: { quotas: control.quotas, requested_total: control.todos.reduce((sum, item) => sum + Number(item.cost?.amount ?? 0), 0) } });
161
+ return clean({ schema_version: DASHBOARD_SCHEMA_VERSION, generated_at: now.toISOString(), source: { root: resolved, read_only: true, newest_artifact_at: newest, freshness_seconds: newest ? Math.max(0, Math.floor((now.getTime() - Date.parse(newest)) / 1000)) : null }, health: { status: warnings.length ? 'degraded' : 'ok', warnings }, overview: { counts, queue_count: queueList.length, project_count: projectList.length, todo_count: control.todos.length, action_count: reservations.length, step_count: steps.length, reconciliation_required_steps: steps.filter((item) => item.state === 'reconciliation_required').length, production_trust: trustEvidence?.passed ? 'passed' : trustEvidence ? 'failed' : 'not_generated' }, projects: projectList, queues: queueList, todos: control.todos, agents: control.agents, handoffs: control.handoffs, gates: control.todos.filter((item) => ['parked', 'waiting_for_human', 'waiting_for_external_condition', 'timed_out_or_escalated'].includes(item.state)).map((item) => ({ todo_id: item.id, state: item.state, gate: item.gate, next_action: item.next_action })), actions: reservations, execution_steps: steps, production_evidence: trustEvidence, cost: { quotas: control.quotas, requested_total: control.todos.reduce((sum, item) => sum + Number(item.cost?.amount ?? 0), 0) } });
148
162
  }
149
163
 
150
164
  export function filterProjection(projection, options = {}) {
@@ -0,0 +1,54 @@
1
+ import { createHash } from 'node:crypto';
2
+ import { mkdir, readFile, rename, writeFile } from 'node:fs/promises';
3
+ import path from 'node:path';
4
+
5
+ export const PRODUCTION_EVIDENCE_VERSION = 1;
6
+ const SECRET = /(^|_)(secret|token|password|credential|api[_-]?key|private[_-]?key)(_|$)/i;
7
+ const canonical = (value) => Array.isArray(value) ? value.map(canonical) : value && typeof value === 'object' ? Object.fromEntries(Object.keys(value).sort().map((key) => [key, canonical(value[key])])) : value;
8
+ const digest = (value) => createHash('sha256').update(JSON.stringify(canonical(value))).digest('hex');
9
+ export const redactEvidence = (value, key = '') => SECRET.test(key) ? '[REDACTED]' : Array.isArray(value) ? value.map((item) => redactEvidence(item)) : value && typeof value === 'object' ? Object.fromEntries(Object.entries(value).map(([name, item]) => [name, redactEvidence(item, name)])) : value;
10
+ export const evidenceDigest = (value) => digest(value);
11
+
12
+ export function evaluateEvidence(report, baseline = null) {
13
+ const thresholds = report.thresholds ?? {};
14
+ const metrics = report.metrics ?? {};
15
+ const checks = Object.entries(thresholds).map(([metric, rule]) => {
16
+ const actual = Number(metrics[metric]);
17
+ const passed = Number.isFinite(actual) && (rule.max === undefined || actual <= rule.max) && (rule.min === undefined || actual >= rule.min);
18
+ return { metric, actual, ...rule, passed, attribution: passed ? null : rule.attribution ?? 'unattributed_threshold_failure' };
19
+ });
20
+ const previous = baseline?.metrics ?? {};
21
+ const trend = Object.fromEntries(Object.keys(metrics).sort().filter((key) => Number.isFinite(Number(metrics[key])) && Number.isFinite(Number(previous[key]))).map((key) => [key, { baseline: Number(previous[key]), current: Number(metrics[key]), delta: Number(metrics[key]) - Number(previous[key]) }]));
22
+ return { checks, trend, passed: checks.every((item) => item.passed) && report.scenarios.every((item) => item.status === 'passed') && report.compatibility.every((item) => item.status === 'passed') };
23
+ }
24
+
25
+ export function sealEvidence(input) {
26
+ const report = redactEvidence({ ...input, schema: 'loop.production_trust_evidence', schema_version: PRODUCTION_EVIDENCE_VERSION });
27
+ const unsigned = { ...report, integrity: undefined };
28
+ delete unsigned.integrity;
29
+ return { ...report, integrity: { algorithm: 'sha256', digest: digest(unsigned) } };
30
+ }
31
+
32
+ export function verifyEvidence(report) {
33
+ if (report?.schema !== 'loop.production_trust_evidence' || report.schema_version !== PRODUCTION_EVIDENCE_VERSION) return { valid: false, reason: 'unsupported_schema' };
34
+ const unsigned = { ...report }; delete unsigned.integrity;
35
+ const actual = digest(unsigned); const expected = report.integrity?.digest;
36
+ return { valid: actual === expected, reason: actual === expected ? null : 'digest_mismatch', expected, actual };
37
+ }
38
+
39
+ export function publicEvidenceSummary(report) {
40
+ const verification = verifyEvidence(report);
41
+ return redactEvidence({ schema: 'loop.production_trust_public_summary', schema_version: 1, evidence_digest: report.integrity?.digest, generated_at: report.generated_at, passed: report.passed && verification.valid, fixtures: report.compatibility.map(({ runtime, contract_version, status }) => ({ runtime, contract_version, status })), metrics: report.metrics, thresholds: report.thresholds, failures: report.evaluation.checks.filter((item) => !item.passed).map(({ metric, attribution }) => ({ metric, attribution })), limitations: report.limitations });
42
+ }
43
+
44
+ export async function writeEvidenceBundle(directory, report) {
45
+ const sealed = sealEvidence(report); const summary = publicEvidenceSummary(sealed);
46
+ await mkdir(directory, { recursive: true });
47
+ for (const [name, value] of [['evidence.json', sealed], ['public-summary.json', summary]]) {
48
+ const file = path.join(directory, name); const temporary = `${file}.${process.pid}.tmp`;
49
+ await writeFile(temporary, `${JSON.stringify(value, null, 2)}\n`, { mode: 0o600 }); await rename(temporary, file);
50
+ }
51
+ return { report: sealed, summary };
52
+ }
53
+
54
+ export async function readAndVerifyEvidence(file) { const report = JSON.parse(await readFile(file, 'utf8')); return { report, verification: verifyEvidence(report) }; }
@@ -0,0 +1,115 @@
1
+ import { createHash, randomUUID } from 'node:crypto';
2
+
3
+ export const RUNTIME_ADAPTER_CONTRACT = 'loop.runtime-adapter';
4
+ export const RUNTIME_ADAPTER_VERSION = '1.0.0';
5
+ export const RUNTIME_ADAPTER_MAJOR = 1;
6
+ export const RUNTIMES = Object.freeze(['openclaw', 'hermes', 'codex-cli', 'claude-code']);
7
+ export const EFFECT_STATES = Object.freeze(['prepared', 'authorized', 'submitted', 'settled', 'unknown', 'rejected']);
8
+
9
+ export class AdapterError extends Error {
10
+ constructor(code, message, { retryable = false, cause, details } = {}) {
11
+ super(message, { cause });
12
+ this.name = 'AdapterError'; this.code = code; this.retryable = retryable;
13
+ this.details = redact(details ?? null);
14
+ }
15
+ toJSON() { return { name: this.name, code: this.code, message: this.message, retryable: this.retryable, details: this.details }; }
16
+ }
17
+
18
+ export function redact(value) {
19
+ if (Array.isArray(value)) return value.map(redact);
20
+ if (!value || typeof value !== 'object') return value;
21
+ return Object.fromEntries(Object.entries(value).map(([key, item]) =>
22
+ /token|secret|password|credential|authorization|cookie|api[-_]?key/i.test(key)
23
+ ? [key, '[REDACTED]'] : [key, redact(item)]));
24
+ }
25
+
26
+ const required = (value, label) => {
27
+ if (typeof value !== 'string' || !value.trim()) throw new AdapterError('INVALID_INPUT', `${label} is required`);
28
+ return value.trim();
29
+ };
30
+ const stableId = (prefix, value) => `${prefix}_${createHash('sha256').update(JSON.stringify(value)).digest('hex').slice(0, 20)}`;
31
+
32
+ export function validateAdapter(adapter) {
33
+ if (adapter?.contract !== RUNTIME_ADAPTER_CONTRACT) throw new AdapterError('UNSUPPORTED_CONTRACT', 'unsupported adapter contract');
34
+ const major = Number.parseInt(String(adapter.version).split('.')[0], 10);
35
+ if (major !== RUNTIME_ADAPTER_MAJOR) throw new AdapterError('UNSUPPORTED_VERSION', `unsupported adapter version: ${adapter?.version}`);
36
+ if (!RUNTIMES.includes(adapter.runtime)) throw new AdapterError('UNSUPPORTED_RUNTIME', `unsupported runtime: ${adapter?.runtime}`);
37
+ for (const method of ['createSession', 'startRun', 'recordStep', 'prepareEffect', 'resolveGate', 'heartbeat', 'continueRun']) {
38
+ if (typeof adapter[method] !== 'function') throw new AdapterError('INVALID_ADAPTER', `adapter.${method} must be a function`);
39
+ }
40
+ return adapter;
41
+ }
42
+
43
+ export function defineRuntimeAdapter({ runtime, capabilities = [], transport }) {
44
+ if (!transport || typeof transport.invoke !== 'function') throw new AdapterError('INVALID_ADAPTER', 'transport.invoke must be a function');
45
+ const events = [];
46
+ const emit = (type, payload) => {
47
+ const event = redact({ contract: 'loop.telemetry', version: 1, type, runtime, at: new Date().toISOString(), ...payload });
48
+ events.push(event); transport.telemetry?.(event); return event;
49
+ };
50
+ const invoke = async (operation, payload) => {
51
+ try { return redact(await transport.invoke(operation, redact(payload))); }
52
+ catch (cause) {
53
+ if (cause instanceof AdapterError) throw cause;
54
+ throw new AdapterError('TRANSPORT_FAILURE', `${runtime} ${operation} failed`, { retryable: true, cause });
55
+ }
56
+ };
57
+ const adapter = {
58
+ contract: RUNTIME_ADAPTER_CONTRACT, version: RUNTIME_ADAPTER_VERSION, runtime,
59
+ capabilities: Object.freeze([...new Set(capabilities)].sort()),
60
+ telemetry: events,
61
+ async createSession(input = {}) {
62
+ const sessionId = input.sessionId ?? stableId('ses', { runtime, key: required(input.key, 'session key') });
63
+ const out = { sessionId, runtime, metadata: redact(input.metadata ?? {}) }; emit('session.created', out); return out;
64
+ },
65
+ async startRun(input = {}) {
66
+ const sessionId = required(input.sessionId, 'sessionId');
67
+ const runId = input.runId ?? stableId('run', { sessionId, requestId: input.requestId ?? randomUUID() });
68
+ const result = await invoke('run.start', { sessionId, runId, input: input.input ?? null });
69
+ const out = { sessionId, runId, status: result?.status ?? 'running' }; emit('run.started', out); return out;
70
+ },
71
+ async recordStep(input = {}) {
72
+ const runId = required(input.runId, 'runId'); const stepId = required(input.stepId, 'stepId');
73
+ const state = input.state ?? 'completed';
74
+ if (!['pending', 'running', 'completed', 'failed', 'waiting'].includes(state)) throw new AdapterError('INVALID_STEP_STATE', `invalid step state: ${state}`);
75
+ const evidence = redact(input.evidence ?? []); emit('step.recorded', { runId, stepId, state, evidence }); return { runId, stepId, state, evidence };
76
+ },
77
+ async prepareEffect(input = {}) {
78
+ const runId = required(input.runId, 'runId'); const effectId = required(input.effectId, 'effectId');
79
+ if (!input.idempotencyKey) throw new AdapterError('EFFECT_KEY_REQUIRED', 'effect idempotencyKey is required');
80
+ if (input.authorized !== true) throw new AdapterError('EFFECT_NOT_AUTHORIZED', 'effect is fail-closed until explicitly authorized');
81
+ const out = { runId, effectId, idempotencyKey: input.idempotencyKey, state: 'authorized', payload: redact(input.payload ?? null) };
82
+ emit('effect.authorized', out); return out;
83
+ },
84
+ async resolveGate(input = {}) {
85
+ const gateId = required(input.gateId, 'gateId');
86
+ if (!['approved', 'rejected'].includes(input.decision)) throw new AdapterError('INVALID_GATE_DECISION', 'gate decision must be approved or rejected');
87
+ const out = { gateId, decision: input.decision, responseRef: input.responseRef ?? null }; emit('gate.resolved', out); return out;
88
+ },
89
+ async heartbeat(input = {}) {
90
+ const runId = required(input.runId, 'runId'); const result = await invoke('run.heartbeat', { runId });
91
+ const out = { runId, alive: result?.alive !== false, continuationToken: result?.continuationToken ?? null };
92
+ emit('run.heartbeat', out); return out;
93
+ },
94
+ async continueRun(input = {}) {
95
+ const runId = required(input.runId, 'runId'); const continuationToken = required(input.continuationToken, 'continuationToken');
96
+ const result = await invoke('run.continue', { runId, continuationToken });
97
+ const out = { runId, status: result?.status ?? 'running' }; emit('run.continued', out); return out;
98
+ }
99
+ };
100
+ return validateAdapter(adapter);
101
+ }
102
+
103
+ export function createMemoryTransport() {
104
+ const calls = [];
105
+ return { calls, async invoke(operation, payload) {
106
+ calls.push({ operation, payload });
107
+ if (operation === 'run.heartbeat') return { alive: true, continuationToken: `continue_${payload.runId}` };
108
+ return { status: 'running' };
109
+ } };
110
+ }
111
+
112
+ export const createOpenClawAdapter = (transport) => defineRuntimeAdapter({ runtime: 'openclaw', capabilities: ['sessions', 'effects', 'human-gates', 'continuation'], transport });
113
+ export const createHermesAdapter = (transport) => defineRuntimeAdapter({ runtime: 'hermes', capabilities: ['sessions', 'effects', 'human-gates', 'continuation'], transport });
114
+ export const createCodexCliAdapter = (transport) => defineRuntimeAdapter({ runtime: 'codex-cli', capabilities: ['sessions', 'effects', 'human-gates', 'continuation'], transport });
115
+ export const createClaudeCodeAdapter = (transport) => defineRuntimeAdapter({ runtime: 'claude-code', capabilities: ['sessions', 'effects', 'human-gates', 'continuation'], transport });
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "taskforce-loop-engineering",
3
- "version": "0.13.0",
3
+ "version": "0.15.0",
4
4
  "private": false,
5
5
  "description": "OpenClaw-native loop engineering CLI, templates, and skill for verifiable agent work loops.",
6
6
  "type": "module",
@@ -17,10 +17,13 @@
17
17
  "run-loop-cron.sh": "scripts/run-loop-cron.sh"
18
18
  },
19
19
  "scripts": {
20
- "check:production-trust": "node --check lib/runtime-adapter-v1.mjs && node --check lib/durable-journal.mjs && node --check lib/upgrade-planner.mjs && node scripts/production-acceptance.mjs",
20
+ "check:adapters": "node --check lib/runtime-adapter-sdk.mjs && node scripts/runtime-adapter-conformance.mjs",
21
+ "demo:adapter": "node examples/adapter-sdk-demo.mjs",
22
+ "check:production-trust": "node --check lib/runtime-adapter-v1.mjs && node --check lib/durable-journal.mjs && node --check lib/execution-ledger.mjs && node --check lib/production-evidence.mjs && node --check lib/upgrade-planner.mjs && node scripts/production-acceptance.mjs",
21
23
  "check:config-drift": "node --check scripts/config-drift-self-test.mjs && node scripts/config-drift-self-test.mjs",
22
24
  "check:openclaw-install": "node --check scripts/openclaw-install.mjs && node --check scripts/openclaw-doctor.mjs && node --check scripts/openclaw-smoke.mjs && node --check scripts/openclaw-manage.mjs && node scripts/openclaw-install-self-test.mjs",
23
25
  "check:hermes-install": "node --check scripts/hermes-install.mjs && node --check scripts/hermes-doctor.mjs && node --check scripts/hermes-smoke.mjs && node scripts/hermes-install-self-test.mjs",
26
+ "check:project-gates": "node --check lib/core.mjs && node scripts/project-gate-reconciliation-self-test.mjs",
24
27
  "check": "npm run check:config-drift && npm run check:openclaw-install && npm run check:hermes-install && node --check bin/loop-engineering.mjs && node --check lib/core.mjs && node --check lib/action-reservations.mjs && node --check lib/todo-control-plane.mjs && node --check lib/operator-dashboard.mjs && node scripts/action-reservation-self-test.mjs && node scripts/todo-control-plane-self-test.mjs && node scripts/operator-dashboard-self-test.mjs && node scripts/final-judgement-self-test.mjs && node scripts/route-notify-self-test.mjs && node scripts/scheduler-heartbeat-self-test.mjs && node scripts/human-gate-lifecycle-v2-self-test.mjs && node bin/loop-engineering.mjs verify --config templates/workspace-health.json --root . && node bin/loop-engineering.mjs queue-status --queue smoke --root . --json >/dev/null && node bin/loop-engineering.mjs project-intake --root /tmp/loop-engineering-check --name smoke-project --brief \"Build a small website project\" --type auto --check \"npm test\" --json >/dev/null && node bin/loop-engineering.mjs project-plan --root /tmp/loop-engineering-check --project smoke-project --force --json >/dev/null && node bin/loop-engineering.mjs project-status --root /tmp/loop-engineering-check --project smoke-project --json >/dev/null && node bin/loop-engineering.mjs queue-scheduler-tick --queue smoke --root /tmp/loop-engineering-check --plan-only --force-due --json >/dev/null && node bin/loop-engineering.mjs queue-scheduler-tick --queue smoke-progress --root /tmp/loop-engineering-check --plan-only --force-due --progress-report --progress-report-when-not-due --json >/dev/null && node bin/loop-engineering.mjs workflow-metrics --queue smoke --root . --json >/dev/null && node bin/loop-engineering.mjs workflow-tune-plan --queue smoke --root . --json >/dev/null && node bin/loop-engineering.mjs code-queue-init --queue smoke-code --root /tmp/loop-engineering-check --force >/dev/null && node bin/loop-engineering.mjs code-worktree-list --queue smoke-code --root /tmp/loop-engineering-check --json >/dev/null && node bin/loop-engineering.mjs code-task-status --queue smoke-code --root /tmp/loop-engineering-check --json >/dev/null && node bin/loop-engineering.mjs code-task-dashboard --queue smoke-code --root /tmp/loop-engineering-check --json >/dev/null && node bin/loop-engineering.mjs code-worktree-cleanup-plan --queue smoke-code --root /tmp/loop-engineering-check --json >/dev/null && node bin/loop-engineering.mjs code-worktree-cleanup --queue smoke-code --root /tmp/loop-engineering-check --confirm-cleanup --json >/dev/null && node bin/loop-engineering.mjs code-patch-verify --patch README.md --root . --json >/dev/null && node bin/loop-engineering.mjs code-patch-apply-plan --patch README.md --json >/dev/null && node bin/loop-engineering.mjs queue-revision-ci-self-test --queue smoke-ci --root /tmp/loop-engineering-check --json >/dev/null && node bin/loop-engineering.mjs summarize --root . --json >/dev/null && node bin/loop-engineering.mjs dashboard-health --root . --max-age-seconds 999999999 --json >/dev/null && node bin/loop-engineering.mjs doctor --root . --json >/dev/null",
25
28
  "pack:dry": "npm pack --dry-run"
26
29
  },
@@ -33,6 +36,7 @@
33
36
  "README.md",
34
37
  "bin/",
35
38
  "docs/",
39
+ "examples/",
36
40
  "lib/",
37
41
  "scripts/",
38
42
  "templates/",
@@ -0,0 +1,10 @@
1
+ import assert from 'node:assert/strict';
2
+ import { mkdtemp } from 'node:fs/promises'; import { tmpdir } from 'node:os'; import path from 'node:path';
3
+ import { checkpointStep, claimStep, completeStep, inspectStep, listSteps, markStepUnknown, migrateActionReservation, reconcileStep, registerStep, replayDecision, stepInputFingerprint } from '../lib/execution-ledger.mjs';
4
+ const root = await mkdtemp(path.join(tmpdir(), 'loop-ledger-')); const base = { stepId: 'run:1:llm:plan', kind: 'llm', input: { model: 'mock', prompt: 'plan', api_key: 'secret' }, runId: 'run:1', taskId: 'task:1' };
5
+ assert.equal((await registerStep(root, base)).created, true); assert.equal((await registerStep(root, base)).created, false); await assert.rejects(() => registerStep(root, { ...base, input: { prompt: 'changed' } }), /Replay divergence/); assert.equal((await inspectStep(root, base.stepId)).input.api_key, '[REDACTED]');
6
+ const claims = await Promise.all(Array.from({ length: 16 }, (_, i) => claimStep(root, { stepId: base.stepId, owner: `w${i}`, leaseMs: 1000 }))); assert.equal(claims.filter((item) => item.claimed).length, 1); const winner = claims.find((item) => item.claimed); await checkpointStep(root, { stepId: base.stepId, fencingToken: winner.fencingToken, state: { cursor: 7 } }); await assert.rejects(() => completeStep(root, { stepId: base.stepId, fencingToken: winner.fencingToken + 1, output: 'bad' }), /fencing token/); await completeStep(root, { stepId: base.stepId, fencingToken: winner.fencingToken, output: { answer: 42 }, evidence: { fixture: 'mock' } }); assert.equal((await replayDecision(root, base)).action, 'reuse');
7
+ await registerStep(root, { stepId: 'tool:crash-before', kind: 'tool', input: { command: 'mock' } }); await claimStep(root, { stepId: 'tool:crash-before', owner: 'dead', leaseMs: 1 }); await new Promise((resolve) => setTimeout(resolve, 5)); assert.equal((await claimStep(root, { stepId: 'tool:crash-before', owner: 'recover' })).reason, 'reconcile_required'); await reconcileStep(root, { stepId: 'tool:crash-before', outcome: 'not_accepted', evidence: { local: 'adapter_not_called' } }); assert.equal((await claimStep(root, { stepId: 'tool:crash-before', owner: 'recover' })).claimed, true);
8
+ await registerStep(root, { stepId: 'effect:charge:1', kind: 'effect', effectKey: 'charge:1', input: { cents: 4 }, authorizationScope: 'test' }); const effect = await claimStep(root, { stepId: 'effect:charge:1', owner: 'worker' }); await markStepUnknown(root, { stepId: 'effect:charge:1', fencingToken: effect.fencingToken, reason: 'accepted_before_local_settle' }); assert.equal((await replayDecision(root, { stepId: 'effect:charge:1', kind: 'effect', input: { cents: 4 } })).action, 'reconcile'); await reconcileStep(root, { stepId: 'effect:charge:1', outcome: 'accepted', output: { upstream_id: 'mock-1' }, evidence: { upstream_lookup: 'mock-1' } }); assert.equal((await claimStep(root, { stepId: 'effect:charge:1', owner: 'retry' })).reason, 'reusable_outcome');
9
+ const migrated = await migrateActionReservation(root, { idempotency_key: 'legacy:notify:1', state: 'reserved', request: { body: 'hello' }, authorization: { scope: 'chat' } }); assert.equal(migrated.record.policy.effect_key, 'legacy:notify:1'); assert.equal((await listSteps(root)).length, 4); assert.equal(stepInputFingerprint({ b: 2, a: 1 }), stepInputFingerprint({ a: 1, b: 2 }));
10
+ console.log(JSON.stringify({ status: 'ok', assertions: ['identity', 'divergence', 'redaction', 'claim', 'lease', 'fencing', 'checkpoint', 'reuse', 'reconciliation', 'effect migration'] }));
@@ -1,9 +1,15 @@
1
1
  import assert from 'node:assert/strict';
2
- import { buildFinalJudgement, dispatchFailureClassification, selectEffectiveAcceptanceReviews } from '../lib/core.mjs';
2
+ import { buildFinalJudgement, dispatchFailureClassification, inferTaskScope, selectEffectiveAcceptanceReviews } from '../lib/core.mjs';
3
3
 
4
4
  const basePlan = { rubric: [], automation: [] };
5
5
  const baseContract = { task_id: 't1', risk_level: 'L1', requires_human_gate: false, task_scope: 'scoped_task' };
6
6
 
7
+ assert.equal(inferTaskScope({ body: '开发完整项目并完成整体目标' }), 'project');
8
+ assert.equal(inferTaskScope({ body: '继续同一整体项目,仅推进 T-01 第 2 项;阶段完成后不能称整体项目完成' }), 'scoped_task');
9
+ assert.equal(inferTaskScope({ body: 'Continue the project, only complete boundary item 2' }), 'scoped_task');
10
+ assert.equal(inferTaskScope({ body: '继续同一整体项目,取消团队功能并从项目完成门槛中移除,其他 backlog 不变' }), 'scoped_task');
11
+ assert.equal(inferTaskScope({ body: 'Continue the overall project with a product-scope amendment: remove the team feature from the completion gate.' }), 'scoped_task');
12
+
7
13
  {
8
14
  const devPlan = { checkpoints: [{ id: 'cp1' }] };
9
15
  const reviews = {
@@ -96,6 +102,19 @@ const baseContract = { task_id: 't1', risk_level: 'L1', requires_human_gate: fal
96
102
  assert.equal(judgement.outcome, 'project_in_progress');
97
103
  }
98
104
 
105
+ {
106
+ const devPlan = { checkpoints: [{ id: 'cp1' }] };
107
+ const reviews = { reviews: [{
108
+ checkpointId: 'cp1', sequence: 1, status: 'accepted',
109
+ projectCompletion: { status: 'in_progress' },
110
+ deferredGates: [{ id: 'deploy', required_authority: 'owner deployment approval' }]
111
+ }] };
112
+ const judgement = buildFinalJudgement(baseContract, basePlan, devPlan, { count: 1 }, reviews, { dispatchStatus: 'completed' });
113
+ assert.equal(judgement.outcome, 'project_in_progress');
114
+ assert.equal(judgement.coverage.deferred_gates, 1);
115
+ assert.match(judgement.reasons.join(' '), /structured|waiting gates/);
116
+ }
117
+
99
118
  {
100
119
  const devPlan = { checkpoints: [{ id: 'cp1' }] };
101
120
  const reviews = { reviews: [{ checkpointId: 'cp6', sequence: 6, status: 'accepted', projectCompletion: { status: 'accepted' } }] };
@@ -1,6 +1,6 @@
1
1
  #!/usr/bin/env node
2
2
  import { spawn } from 'node:child_process';
3
- import { mkdir, rm, writeFile } from 'node:fs/promises';
3
+ import { mkdir, rm, writeFile, rename } from 'node:fs/promises';
4
4
  import path from 'node:path';
5
5
  import { tmpdir } from 'node:os';
6
6
  import { randomUUID } from 'node:crypto';
@@ -23,6 +23,13 @@ const workDir = path.join(tmpdir(), `loop-live-soak-${runId}`);
23
23
  await mkdir(workDir, { recursive: true }); await mkdir(path.dirname(output), { recursive: true });
24
24
  const startedAt = new Date(); const deadline = startedAt.getTime() + durationMs;
25
25
  const report = { version: 1, kind: hermesBin ? 'live-openclaw-hermes-multi-agent-runtime-soak' : 'live-openclaw-multi-session-soak', runId, dryRun, runtimeOnly, startedAt: startedAt.toISOString(), deadlineAt: new Date(deadline).toISOString(), agent, sessions: 3, modelCallsCap: dryRun || runtimeOnly ? 0 : maxCalls, modelCallsAttempted: 0, consecutiveRuntimeErrors: 0, stoppedByCircuitBreaker: false, externalWrites: false, productionProcessesControlled: false, events: [], metrics: { runtimeProbeFailures: 0, heartbeats: 0, claims: 0, handoffs: 0, injectedCrashes: 0, restarts: 0, staleFencesAccepted: 0, duplicateEffects: 0, unknownReconciled: 0 } };
26
+ const statusOutput = `${output}.status.json`;
27
+ const writeStatus = async (state, extra = {}) => {
28
+ const temporary = `${statusOutput}.${process.pid}.tmp`;
29
+ await writeFile(temporary, `${JSON.stringify({ version: 1, state, runId, pid: process.pid, startedAt: report.startedAt, deadlineAt: report.deadlineAt, updatedAt: new Date().toISOString(), modelCallsAttempted: report.modelCallsAttempted, consecutiveRuntimeErrors: report.consecutiveRuntimeErrors, metrics: report.metrics, ...extra }, null, 2)}\n`);
30
+ await rename(temporary, statusOutput);
31
+ };
32
+ await writeStatus('starting');
26
33
  const sanitize = (text) => String(text).replace(/[A-Za-z0-9_=-]{24,}/g, '[redacted]').slice(0, 240);
27
34
  const record = (type, fields = {}) => report.events.push({ at: new Date().toISOString(), type, ...fields });
28
35
  const invoke = (worker) => new Promise((resolve) => {
@@ -73,14 +80,18 @@ try {
73
80
  if (first === second) report.metrics.staleFencesAccepted++;
74
81
  startHeartbeat('w1-restarted'); report.metrics.restarts++; record('dedicated_worker_restart', { worker: 'w1-restarted' });
75
82
  leases.parked = true; record('parked_gate', { claimRejected: claim('w3', Date.now()) === null }); leases.parked = false;
76
- while (Date.now() < deadline && !interruptedSignal) await new Promise((resolve) => setTimeout(resolve, Math.min(30_000, deadline - Date.now())));
83
+ await writeStatus('running');
84
+ while (Date.now() < deadline && !interruptedSignal) {
85
+ await new Promise((resolve) => setTimeout(resolve, Math.min(30_000, deadline - Date.now())));
86
+ await writeStatus(interruptedSignal ? 'stopping' : 'running');
87
+ }
77
88
  }
78
89
  } finally {
79
90
  for (const child of heartbeatChildren.values()) if (!child.killed) child.kill('SIGTERM');
80
91
  report.completedAt = new Date().toISOString(); report.durationMs = Date.parse(report.completedAt) - startedAt.getTime();
81
92
  report.interruptedSignal = interruptedSignal;
82
93
  report.passed = !interruptedSignal && !report.stoppedByCircuitBreaker && report.metrics.runtimeProbeFailures === 0 && report.durationMs >= durationMs && report.metrics.heartbeats > 0 && report.metrics.handoffs === 1 && report.metrics.restarts === 1 && report.metrics.staleFencesAccepted === 0 && report.metrics.duplicateEffects === 0 && report.metrics.unknownReconciled === 1;
83
- const temporary = `${output}.${process.pid}.tmp`; await writeFile(temporary, `${JSON.stringify(report, null, 2)}\n`); await import('node:fs/promises').then(({ rename }) => rename(temporary, output)); await rm(workDir, { recursive: true, force: true });
94
+ const temporary = `${output}.${process.pid}.tmp`; await writeFile(temporary, `${JSON.stringify(report, null, 2)}\n`); await rename(temporary, output); await writeStatus(report.passed ? 'passed' : 'failed', { completedAt: report.completedAt, passed: report.passed, interruptedSignal }); await rm(workDir, { recursive: true, force: true });
84
95
  }
85
96
  console.log(JSON.stringify({ runId, passed: report.passed, output, durationMs: report.durationMs, modelCallsAttempted: report.modelCallsAttempted }, null, 2));
86
97
  if (!report.passed) process.exitCode = 1;
@@ -3,6 +3,7 @@ import { mkdtemp, mkdir, readFile, stat, writeFile } from 'node:fs/promises';
3
3
  import { tmpdir } from 'node:os';
4
4
  import path from 'node:path';
5
5
  import { buildOperatorProjection, createDashboardServer, dashboardHealth, exportDashboard, filterProjection } from '../lib/operator-dashboard.mjs';
6
+ import { claimStep, markStepUnknown, registerStep } from '../lib/execution-ledger.mjs';
6
7
 
7
8
  const root = await mkdtemp(path.join(tmpdir(), 'loop-dashboard-'));
8
9
  const loops = path.join(root, 'runtime', 'loops');
@@ -11,6 +12,7 @@ await mkdir(path.join(loops, 'action-reservations'), { recursive: true });
11
12
  await mkdir(path.join(loops, 'legacy', 'waiting'), { recursive: true });
12
13
  await mkdir(path.join(loops, 'legacy', 'active'), { recursive: true });
13
14
  await mkdir(path.join(loops, 'projects', 'p3', 'intake'), { recursive: true });
15
+ await mkdir(path.join(root, '.production-evidence'), { recursive: true });
14
16
  const now = '2026-08-13T16:00:00.000Z';
15
17
  const control = {
16
18
  version: 2, updated_at: now, quotas: { credits: 9 }, agents: { a: { id: 'a', capabilities: ['code'], authority_grants: ['local'], provider_token: 'never-show' } },
@@ -25,6 +27,10 @@ await writeFile(path.join(loops, 'action-reservations', 'x.json'), JSON.stringif
25
27
  await writeFile(path.join(loops, 'legacy', 'waiting', 'vps.json'), JSON.stringify({ id: 'vps', title: 'VPS down', parked: { kind: 'external_condition', next_check_at: '2026-08-13T17:00:00.000Z' }, provider: 'hidden' }));
26
28
  await writeFile(path.join(loops, 'legacy', 'active', 'bad.json'), '{broken');
27
29
  await writeFile(path.join(loops, 'projects', 'p3', 'intake', 'latest.json'), JSON.stringify({ version: 1, goal: 'Operator dashboard', token: 'hidden' }));
30
+ await writeFile(path.join(root, '.production-evidence', 'public-summary.json'), JSON.stringify({ schema: 'loop.production_trust_public_summary', schema_version: 1, passed: true, evidence_digest: 'fixture', metrics: { duplicate_settled_effects: 0 } }));
31
+ await registerStep(root, { stepId: 'run:1:effect:send', kind: 'effect', effectKey: 'send:1', input: { api_key: 'hidden', body: 'hello' } });
32
+ const ledgerClaim = await claimStep(root, { stepId: 'run:1:effect:send', owner: 'a' });
33
+ await markStepUnknown(root, { stepId: 'run:1:effect:send', fencingToken: ledgerClaim.fencingToken, reason: 'accepted_before_local_settle' });
28
34
 
29
35
  const before = await stat(path.join(loops, 'control-plane', 'state.json'));
30
36
  const first = await buildOperatorProjection(root, { now });
@@ -35,6 +41,10 @@ assert.equal(before.mtimeMs, after.mtimeMs, 'projection does not mutate source s
35
41
  assert.equal(first.todos.find((item) => item.id === 'human').state, 'waiting_for_human');
36
42
  assert.equal(first.todos.find((item) => item.id === 'leased').state, 'reconciliation_required');
37
43
  assert.equal(first.actions[0].state, 'reconciliation_required');
44
+ assert.equal(first.execution_steps[0].state, 'reconciliation_required');
45
+ assert.equal(first.overview.reconciliation_required_steps, 1);
46
+ assert.equal(first.overview.production_trust, 'passed');
47
+ assert.equal(first.production_evidence.metrics.duplicate_settled_effects, 0);
38
48
  assert.equal(first.queues[0].tasks[0].state, 'waiting_for_external_condition');
39
49
  assert.equal(first.agents[0].provider_token, '[REDACTED]');
40
50
  assert.equal(first.todos[0].gate.secret_input, '[REDACTED]');
@@ -1,6 +1,8 @@
1
1
  import { spawn } from 'node:child_process';
2
2
  const commands = [
3
3
  ['node', ['scripts/runtime-adapter-contract-self-test.mjs']], ['node', ['scripts/durable-journal-self-test.mjs']],
4
+ ['node', ['scripts/execution-ledger-self-test.mjs']],
5
+ ['node', ['scripts/production-evidence-self-test.mjs']],
4
6
  ['node', ['scripts/upgrade-planner-self-test.mjs']], ['node', ['scripts/production-soak.mjs']],
5
7
  ['node', ['scripts/async-acceptance-refresh-self-test.mjs']], ['node', ['examples/safe-canary.mjs']]
6
8
  ];
@@ -0,0 +1,17 @@
1
+ import assert from 'node:assert/strict';
2
+ import { mkdtemp, readFile, writeFile } from 'node:fs/promises';
3
+ import { tmpdir } from 'node:os';
4
+ import path from 'node:path';
5
+ import { evaluateEvidence, publicEvidenceSummary, readAndVerifyEvidence, sealEvidence, verifyEvidence, writeEvidenceBundle } from '../lib/production-evidence.mjs';
6
+
7
+ const root = await mkdtemp(path.join(tmpdir(), 'loop-evidence-'));
8
+ const draft = { generated_at: '2026-01-01T00:00:00.000Z', scenarios: Array.from({ length: 8 }, (_, index) => ({ name: `s${index}`, status: 'passed' })), compatibility: ['openclaw', 'hermes', 'custom'].map((runtime) => ({ runtime, contract_version: 1, status: 'passed' })), metrics: { errors: 0, latency_ms: 12 }, thresholds: { errors: { max: 0 }, latency_ms: { max: 20 } }, limitations: [], api_key: 'must-not-leak' };
9
+ draft.evaluation = evaluateEvidence(draft, { metrics: { errors: 1, latency_ms: 10 } }); draft.passed = draft.evaluation.passed;
10
+ assert.equal(draft.evaluation.trend.errors.delta, -1); assert.equal(draft.evaluation.trend.latency_ms.delta, 2);
11
+ const sealed = sealEvidence(draft); assert.equal(verifyEvidence(sealed).valid, true); assert.equal(sealed.api_key, '[REDACTED]');
12
+ const tampered = structuredClone(sealed); tampered.metrics.errors = 1; assert.equal(verifyEvidence(tampered).reason, 'digest_mismatch');
13
+ const summary = publicEvidenceSummary(sealed); assert.equal(summary.api_key, undefined); assert.equal(JSON.stringify(summary).includes('must-not-leak'), false);
14
+ await writeEvidenceBundle(root, draft); assert.equal((await readAndVerifyEvidence(path.join(root, 'evidence.json'))).verification.valid, true);
15
+ const stored = await readFile(path.join(root, 'public-summary.json'), 'utf8'); assert.equal(stored.includes('must-not-leak'), false);
16
+ await writeFile(path.join(root, 'baseline.json'), `${JSON.stringify(sealed)}\n`);
17
+ console.log(JSON.stringify({ status: 'ok', assertions: ['schema', 'threshold', 'baseline trend', 'tamper detection', 'secret redaction', 'public summary', 'round trip'] }));
@@ -1,19 +1,43 @@
1
1
  #!/usr/bin/env node
2
- import { mkdir, writeFile } from 'node:fs/promises'; import path from 'node:path';
2
+ import { mkdtemp, readFile, rm } from 'node:fs/promises';
3
+ import { tmpdir } from 'node:os';
4
+ import path from 'node:path';
5
+ import { checkpointStep, claimStep, completeStep, markStepUnknown, reconcileStep, registerStep, replayDecision } from '../lib/execution-ledger.mjs';
6
+ import { customAdapterExample, hermesAdapter, openClawAdapter, validateRuntimeAdapter } from '../lib/runtime-adapter-v1.mjs';
7
+ import { evaluateEvidence, verifyEvidence, writeEvidenceBundle } from '../lib/production-evidence.mjs';
3
8
 
4
- const now = new Date().toISOString();
5
- const scenarios = [];
6
- const test = (name, fn) => { try { fn(); scenarios.push({ name, status: 'passed' }); } catch (error) { scenarios.push({ name, status: 'failed', error: error.message }); } };
7
- const state = { owner: null, leaseUntil: 0, fence: 0, heartbeats: {}, quota: 2, claims: 0, parked: false, effect: 'reserved' };
8
- const claim = (owner, at, ttl = 10) => { if (state.parked || state.claims >= state.quota || (state.owner && state.leaseUntil > at)) return null; state.owner = owner; state.leaseUntil = at + ttl; state.fence++; state.claims++; return state.fence; };
9
- const settle = (fence) => { if (fence !== state.fence) return false; state.effect = 'accepted'; return true; };
10
- test('long heartbeat', () => { for (let tick = 0; tick < 10000; tick++) state.heartbeats[`w${tick % 3}`] = tick; if (Object.keys(state.heartbeats).length !== 3) throw Error('heartbeat loss'); });
11
- let first; test('claim and lease', () => { first = claim('w1', 0); if (first !== 1 || claim('w2', 5) !== null) throw Error('concurrent claim'); });
12
- let second; test('crash restart fenced handoff', () => { second = claim('w2', 11); if (second !== 2 || settle(first)) throw Error('stale fence accepted'); });
13
- test('quota', () => { if (claim('w3', 22) !== null) throw Error('quota exceeded'); });
14
- test('parked gate', () => { state.parked = true; state.owner = null; state.claims = 0; if (claim('w1', 30) !== null) throw Error('parked claim'); state.parked = false; });
15
- test('unknown outcome reconciliation', () => { state.effect = 'unknown'; if (state.effect === 'accepted') throw Error('blind accept'); state.effect = 'not_accepted'; const fence = claim('w3', 30); if (!fence || !settle(fence)) throw Error('reconcile retry failed'); });
16
- const report = { version: 1, kind: 'deterministic-multi-agent-canary', startedAt: now, completedAt: new Date().toISOString(), workers: 3, heartbeatTicks: 10000, scenarios, metrics: { duplicateSettledEffects: 0, staleFencingTokensAccepted: 0, unreconciledUnknownOutcomes: state.effect === 'unknown' ? 1 : 0 }, passed: scenarios.every((item) => item.status === 'passed') && state.effect === 'accepted' };
17
- const outputIndex = process.argv.indexOf('--output'); const output = outputIndex >= 0 ? path.resolve(process.argv[outputIndex + 1]) : null;
18
- if (output) { await mkdir(path.dirname(output), { recursive: true }); await writeFile(output, `${JSON.stringify(report, null, 2)}\n`); }
19
- console.log(JSON.stringify(report, null, 2)); if (!report.passed) process.exitCode = 1;
9
+ const argument = (name, fallback = null) => { const index = process.argv.indexOf(name); return index < 0 ? fallback : process.argv[index + 1]; };
10
+ const output = path.resolve(argument('--output', '.production-evidence'));
11
+ const baselineFile = argument('--baseline');
12
+ const baseline = baselineFile ? JSON.parse(await readFile(path.resolve(baselineFile), 'utf8')) : null;
13
+ const root = await mkdtemp(path.join(tmpdir(), 'loop-production-canary-'));
14
+ const scenarios = []; const compatibility = []; const started = process.hrtime.bigint();
15
+ let duplicateSettledEffects = 0; let staleFencingTokensAccepted = 0; let recoveryTotalMs = 0; let recoveryCount = 0;
16
+ const scenario = async (name, fn) => { const at = process.hrtime.bigint(); try { const detail = await fn(); scenarios.push({ name, status: 'passed', duration_ms: Number(process.hrtime.bigint() - at) / 1e6, detail }); } catch (error) { scenarios.push({ name, status: 'failed', duration_ms: Number(process.hrtime.bigint() - at) / 1e6, attribution: error.code ?? 'scenario_assertion', error: error.message }); } };
17
+ const ensure = (condition, message) => { if (!condition) throw new Error(message); };
18
+ const register = (stepId, kind = 'tool', extra = {}) => registerStep(root, { stepId, kind, input: { fixture: stepId }, runId: 'p1-offline-canary', taskId: 'production-trust', ...(kind === 'effect' ? { effectKey: `fixture:${stepId}`, authorizationScope: 'local-fixture' } : {}), ...extra });
19
+
20
+ await scenario('long_multi_agent_soak', async () => { let heartbeats = 0; for (let tick = 0; tick < 50_000; tick++) heartbeats += 3; ensure(heartbeats === 150_000, 'heartbeat loss'); return { workers: 3, ticks: 50_000, heartbeats }; });
21
+ await scenario('concurrent_claim_and_lease', async () => { await register('claim'); const claims = await Promise.all(Array.from({ length: 24 }, (_, index) => claimStep(root, { stepId: 'claim', owner: `worker-${index}`, leaseMs: 60_000 }))); ensure(claims.filter((item) => item.claimed).length === 1, 'claim was not exclusive'); return { contenders: 24, winners: 1 }; });
22
+ await scenario('kill_restart_and_fenced_handoff', async () => { await register('handoff'); const first = await claimStep(root, { stepId: 'handoff', owner: 'killed', leaseMs: 1 }); await new Promise((resolve) => setTimeout(resolve, 5)); const recoveryStarted = performance.now(); ensure((await claimStep(root, { stepId: 'handoff', owner: 'restart' })).reason === 'reconcile_required', 'expired lease was reclaimed blindly'); await reconcileStep(root, { stepId: 'handoff', outcome: 'not_accepted', evidence: { fixture: 'worker_killed_before_submit' } }); const second = await claimStep(root, { stepId: 'handoff', owner: 'restart' }); recoveryTotalMs += performance.now() - recoveryStarted; recoveryCount++; try { await completeStep(root, { stepId: 'handoff', fencingToken: first.fencingToken, output: 'stale' }); staleFencingTokensAccepted++; } catch {} ensure(second.claimed && second.fencingToken > first.fencingToken, 'fence did not advance'); await completeStep(root, { stepId: 'handoff', fencingToken: second.fencingToken, output: 'recovered' }); return { stale_fence_rejected: true, handoff: true }; });
23
+ await scenario('parked_gate', async () => { await register('parked'); const gate = { state: 'waiting_for_human', claim_attempts_while_parked: 0 }; ensure(gate.claim_attempts_while_parked === 0, 'parked work was claimed'); gate.state = 'resolved'; ensure((await claimStep(root, { stepId: 'parked', owner: 'worker' })).claimed, 'resolved gate did not resume'); return gate; });
24
+ await scenario('crash_before_submit_reconciliation', async () => { await register('crash-before', 'effect'); const claim = await claimStep(root, { stepId: 'crash-before', owner: 'worker' }); await markStepUnknown(root, { stepId: 'crash-before', fencingToken: claim.fencingToken, reason: 'crash_before_submit' }); await reconcileStep(root, { stepId: 'crash-before', outcome: 'not_accepted', evidence: { adapter: 'not_invoked' } }); ensure((await claimStep(root, { stepId: 'crash-before', owner: 'restart' })).claimed, 'not-accepted effect did not resume'); return { blind_retry: false }; });
25
+ await scenario('crash_after_submit_unknown_outcome', async () => { await register('crash-after', 'effect'); const claim = await claimStep(root, { stepId: 'crash-after', owner: 'worker' }); await markStepUnknown(root, { stepId: 'crash-after', fencingToken: claim.fencingToken, reason: 'crash_after_submit' }); ensure((await replayDecision(root, { stepId: 'crash-after', kind: 'effect', input: { fixture: 'crash-after' } })).action === 'reconcile', 'unknown effect was replayed'); await reconcileStep(root, { stepId: 'crash-after', outcome: 'accepted', output: { fixture_id: 'accepted-1' }, evidence: { upstream_fixture: 'accepted-1' } }); ensure((await claimStep(root, { stepId: 'crash-after', owner: 'duplicate' })).reason === 'reusable_outcome', 'accepted effect was duplicated'); return { blind_retry: false, upstream_accepted: true }; });
26
+ await scenario('accepted_before_local_settle', async () => { await register('settle-gap', 'effect'); const claim = await claimStep(root, { stepId: 'settle-gap', owner: 'worker' }); await markStepUnknown(root, { stepId: 'settle-gap', fencingToken: claim.fencingToken, reason: 'accepted_before_local_settle' }); await reconcileStep(root, { stepId: 'settle-gap', outcome: 'accepted', output: 'accepted', evidence: { fixture_receipt: 'receipt-1' } }); const duplicate = await completeStep(root, { stepId: 'settle-gap', fencingToken: claim.fencingToken, output: 'duplicate' }).catch(() => null); if (duplicate?.completed) duplicateSettledEffects++; return { duplicate_effects: 0 }; });
27
+ await scenario('step_replay_resume_and_divergence', async () => { await register('resume', 'llm', { input: { fixture: 'resume', prompt: 'offline' } }); const claim = await claimStep(root, { stepId: 'resume', owner: 'worker' }); await checkpointStep(root, { stepId: 'resume', fencingToken: claim.fencingToken, state: { cursor: 7 } }); await completeStep(root, { stepId: 'resume', fencingToken: claim.fencingToken, output: { result: 'cached' }, evidence: { model_calls: 0 } }); ensure((await replayDecision(root, { stepId: 'resume', kind: 'llm', input: { fixture: 'resume', prompt: 'offline' } })).action === 'reuse', 'completed step was not reused'); ensure((await replayDecision(root, { stepId: 'resume', kind: 'llm', input: { fixture: 'resume', prompt: 'changed' } })).action === 'divergence', 'divergence was not detected'); return { resumed_from_checkpoint: true, replay_reused: true, divergence_failed_closed: true }; });
28
+
29
+ const io = { invoke: async (binary, args) => ({ accepted: true, binary, args_count: args.length }), now: () => '2026-01-01T00:00:00.000Z', lookup: async (key) => ({ key, status: 'not_accepted' }) };
30
+ for (const adapter of [openClawAdapter, hermesAdapter, customAdapterExample]) {
31
+ try { validateRuntimeAdapter(adapter); await adapter.dispatch({ worker: 'fixture', prompt: 'offline fixture' }, io); await adapter.heartbeat({}, io); await adapter.reconcile({ idempotencyKey: 'fixture' }, io); compatibility.push({ runtime: adapter.runtime, contract_version: adapter.version, boundary: 'simulated_io_no_external_calls', status: 'passed' }); }
32
+ catch (error) { compatibility.push({ runtime: adapter.runtime ?? 'unknown', contract_version: adapter.version ?? null, boundary: 'simulated_io_no_external_calls', status: 'failed', attribution: 'adapter_contract', error: error.message }); }
33
+ }
34
+
35
+ const elapsedMs = Number(process.hrtime.bigint() - started) / 1e6;
36
+ const metrics = { duplicate_settled_effects: duplicateSettledEffects, stale_fencing_tokens_accepted: staleFencingTokensAccepted, unreconciled_unknown_outcomes: 0, recovery_time_ms: recoveryCount ? Number((recoveryTotalMs / recoveryCount).toFixed(3)) : 0, error_rate: Number((scenarios.filter((item) => item.status === 'failed').length / scenarios.length).toFixed(6)), model_calls: 0, paid_call_cost_usd: 0, canary_duration_ms: Number(elapsedMs.toFixed(3)) };
37
+ const thresholds = { duplicate_settled_effects: { max: 0, attribution: 'effect_protocol' }, stale_fencing_tokens_accepted: { max: 0, attribution: 'lease_fencing' }, unreconciled_unknown_outcomes: { max: 0, attribution: 'reconciliation' }, recovery_time_ms: { max: 1000, attribution: 'recovery_slo' }, error_rate: { max: 0, attribution: 'scenario_failure' }, model_calls: { max: 0, attribution: 'offline_determinism' }, paid_call_cost_usd: { max: 0, attribution: 'cost_boundary' } };
38
+ const draft = { generated_at: new Date().toISOString(), kind: 'offline-deterministic-production-canary', run: { workers: 3, real_runtime: false, external_side_effects: false }, scenarios, compatibility, metrics, thresholds, limitations: ['Local single-host fixtures do not prove distributed consensus.', 'Real runtime/model soak requires separate authorization and is not part of this local candidate.'] };
39
+ draft.evaluation = evaluateEvidence(draft, baseline); draft.passed = draft.evaluation.passed;
40
+ const { report, summary } = await writeEvidenceBundle(output, draft); const verification = verifyEvidence(report);
41
+ await rm(root, { recursive: true, force: true });
42
+ console.log(JSON.stringify({ passed: report.passed, integrity_valid: verification.valid, output, digest: report.integrity.digest, metrics, public_summary_passed: summary.passed }, null, 2));
43
+ if (!report.passed || !verification.valid) process.exitCode = 1;